Release notes

v0.2.20

Updates

  • Add MKV as valid file extension in AudioContainer

  • Add overlay method in AudioContainer

  • Add balancing_mode parameter to validation_files_balanced method in SoundEventDataset

  • Add dataset field to MetaDataItem and add dataset field related properties to MetaDataContainer

  • Add identifier count reporting for tags in MetaDataContainer

  • Add map_tags method in MetaDataContainer

  • Add TAUUrbanAcousticScenes_2022_Mobile_EvaluationSet dataset

  • Update unique_source_labels method in MetaDataContainer to be more efficient

  • Update model_summary_string function to work with latest Keras versions

Bug fixes

  • Fix field override in non_hashable_fields method in AppParameterContainer

  • Fix process_meta_container in AcousticSceneDataset to retain container filename

  • Fix is_jupyter function to work when IPython is not installed

  • Fix plot in AudioContainer method to use figsize parameter for dual plots

v0.2.19

Updates

  • Add TAUUrbanAcousticScenes_2022_Mobile_DevelopmentSet

  • Add active_scenes and active_events parameter for Datasets class to select scene and event classes.

  • Add event activity and inactivity calculation for MetaDataContainer

  • Add TorchOpenL3Extractor and TorchOpenL3ExtractorProcessor

  • Add float5 and float6 value types to formatted_value method in FancyStringifier.

  • Add valid content type ‘all’ for RemoteFile

Bug fixes

  • Fix unique_files in MetaDataContainer to be more efficient with large number of files.

  • Fix check_metadata in Dataset to be more efficient with large number of files.

  • Fix field ‘event_label’ processing to be robust for non-string values in MetaDataItem

  • Fix data loading to be robust for empty files in DictContainer

  • Fix ‘float1_percentage+ci’, ‘float2_percentage+ci’, ‘float3_percentage+ci’, and ‘float4_percentage+ci’ value types in FancyStringifier to have fallback data types if values do not have full data.

v0.2.18

Updates

  • Add TAUUrbanAudioVisualScenes_2021_EvaluationSet class TAUUrbanAcousticScenes_2021_Mobile_EvaluationSet

Bug fixes

  • Fix normalize method in AudioContainer to work with multi and single channel audio.

  • Fix pack method in DatasetPacker correctly identify changed files and trigger package regeneration

  • Fix TAUUrbanAcousticScenes_2021_Mobile_EvaluationSet to correctly extract zip-packages

v0.2.17

Updates

  • Add TensorFlow keras utilities (tf.keras)

  • Add get_media_duration and merge_media_files functions

  • Add filename_audio and filename_video properties to MetaDataItem class

  • Add TAUUrbanAudioVisualScenes_2021_DevelopmentSet class

  • Update travis tests: Python 2.7 tests are dropped, only Python 3.X tests are used

Bug fixes

  • Fix base_path removal in DatasetPacker

v0.2.16

Updates

  • Add scene_labels for TAUUrbanAcousticScenes_2020_Mobile_EvaluationSet and TAUUrbanAcousticScenes_2020_3Class_EvaluationSet.

  • Update SubmissionChecker

v0.2.15

Bug fixes

  • Fix all_data fold handling in TAUUrbanAcousticScenes_2020_Mobile_EvaluationSet and TAUUrbanAcousticScenes_2020_3Class_EvaluationSet to be uniform.

v0.2.14

Updates

  • Update PIP package

v0.2.13

New features

  • TAUUrbanAcousticScenes_2020_Mobile_EvaluationSet and TAUUrbanAcousticScenes_2020_3Class_EvaluationSet datasets

v0.2.12

Updates

  • Update TAUUrbanAcousticScenes_2020_Mobile_DevelopmentSet to use dataset version 2.0

v0.2.11

New features

  • Add OpenL3Extractor, EdgeL3Extractor, and EmbeddingExtractor feature extractor classes

  • Add OpenL3ExtractorProcessor and EdgeL3ExtractorProcessor processors

  • Add TAUUrbanAcousticScenes_2020_Mobile_DevelopmentSet and TAUUrbanAcousticScenes_2020_3Class_DevelopmentSet datasets

  • Add get_audio_info function to allow fetching audio file information without reading full file.

  • Add MP3 audio example file

Updates

  • Update AudioContainer constructor to allow initialization with multi-channel audio data in form of list of audio data vectors.

  • Update load method in AudioContainer to have parameter auto_trimming to automatically trim stop parameter to audio file length

  • Update load method in AudioContainer to check start and stop parameters against actual audio file duration

  • Update AudioContainer to store channel_labels, update plot_wave and plot_spec methods with channel_labels as well.

  • Update plot_wave method in AudioContainer to support max_sr parameter and different color per channel.

  • Update plot method in AudioContainer to plot both waveform and spectrogram at the same time (dual plotting mode)

  • Update segments method in AudioContainer to support active_segments

  • Update FancyLogger, FancyHTMLPrinter, and FancyStringifier to accumulate row value when using row method, and add row_sum and row_average methods

  • Update setup_keras to suppress TensorFlow warnings

  • Update debug_packages method of Dataset class to show more information about local files

  • Update FileMixin to allow overriding valid_format through constructor parameter

  • Update code to support Librosa 0.7.0

Bug fixes

  • Fix MetaDataContainer sorting to work with numeric filenames

  • Fix get_byte_string to work with small values

  • Fix filename handling in FeatureRepository when dict of filenames is used

  • Fix collapse_probabilities_windowed method in ProbabilityEncoder to accept arrays of probabilities

  • Fix example system sed_gmm.py to work with current version

v0.2.10

Bug fixes

  • Fix cross-validation data loading for datasets without reference meta data in load_crossvalidation_data method of Dataset class

v0.2.9

New features

  • Add TAUUrbanAcousticScenes_2019_EvaluationSet, TAUUrbanAcousticScenes_2019_Mobile_EvaluationSet, and TAUUrbanAcousticScenes_2019_Openset_EvaluationSet datasets.

v0.2.8

New features

  • Add TAUUrbanAcousticScenes_2019_LeaderboardSet, TAUUrbanAcousticScenes_2019_Mobile_LeaderboardSet, and TAUUrbanAcousticScenes_2019_Openset_LeaderboardSet datasets.

  • Add is_jupyter function to detect if code is running inside jupyter

  • Add shorten method in Path to shorten long paths for visualization purpose

  • Add FancyHTMLStringifier, FancyHTMLPrinter classes for HTML output

  • Add plot method in DataArrayContainer

  • Add plot method in Normalizer

Updates

  • Update YAML serialization to use yaml.FullLoader

  • formatted_value method in FancyStringifier to be static

  • Refactor printing methods in containers to allow automatic output mode switching between HTML (Jypyter) and string (console)

  • Update data printing mechanism for containers

  • Update plot methods API to include figsize parameter

  • Update default parameters in plot method in AudioContainer (color bar is hidden by default)

  • Update error messages in AudioContainer to be more informative

  • Update load method in MetaDataContainer to support additional row formats

  • Update feature_extractor_list method to have option to return string or display (print to console or print as HTML output in Jupyter)

  • Update dataset_list to use table layout and add option to return string or display (print to console or print as HTML output in Jupyter)

  • Update to_string method in MetaDataContainer with option show_info to control what data is print

  • Update API for methods show and log in Dataset to include show_meta parameter and mode parameter to control output format

  • Update printing validation_files_balanced method in AcousticSceneDataset to support different output modes (print to console or print as HTML output in Jupyter)

  • Update ProgressLoggerCallback to include show_timing parameter and notebook output type

  • Update StasherCallback with to_string and show

  • Update printing inside setup_keras function

  • Update model_summary_string function with new parameters (show_parameters and display)

  • Update plot method in DataMatrix2DContainer with xlabel and ylabel parameters

  • Update plot method in BinaryMatrix2DContainer with panel_title_position parameters

  • Update usage of tqdm library in Dataset to allow locally progress bar disable/enable

Bug fixes

  • Fix single channel audio plotting in AudioContainer

v0.2.7

Updates

  • Update TAUUrbanAcousticScenes_2019_Mobile_DevelopmentSet, and TAUUrbanAcousticScenes_2019_Openset_DevelopmentSet datasets.

v0.2.6

New features

  • Add TAUUrbanAcousticScenes_2019_DevelopmentSet, TAUUrbanAcousticScenes_2019_Mobile_DevelopmentSet, and TAUUrbanAcousticScenes_2019_Openset_DevelopmentSet datasets.

  • Add OneHotEncoder and OneHotEncodingProcessor to allow unknown labels.

  • Add automatic meta data check ups in datasets classes, and parameter to control it.

  • Add AudioSequencingProcessor

  • Add feature_extractor_list to show all available feature extractors classes, and add description to all feature extraction classes.

Updates

  • Update debug_packages method to allow better control which part of package_list is checked: remote or local.

  • Update data_collector to have generic data axis handling.

  • Update load method in ListDictContainer to skip empty rows in CSV files.

  • Update save method in ListDictContainer for TXT and CSV to avoid extra empty lines under Windows.

  • Update save method in MetaDataContainer for TXT and CSV to avoid extra empty lines under Windows.

  • Update relative_to_absolute_path and absolute_to_relative_path to give more informative error messages.

  • Update EventRollEncodingProcessor to support pad_length parameter.

  • Update unit tests to be cross-platform compatible (Linux / Windows)

  • Update SuppressStdoutAndStderr to be more robust

  • Update MetaDataItem to keep filename field to be posix path when relative path is used.

  • Update dtypes to be compatible with numpy v1.14

  • Update setup_keras to warn when GPU was not found.

  • Update model_summary_string to show activation function of the output layer.

  • Update all processors, encoders, and manipulators have __call__ magic class method.

Bug fixes

  • Fix delimiter detection in load method in MetaDataContainer

  • Fix MetaDataItem to better handle empty fields (onset, offset, and event_label).

  • Fix how validation_split and validation_files_dataset method uses training_meta parameter.

v0.2.5

New features

  • Add SoundDataset base class.

  • Add feature_extractor_factory to get feature extractor class based on feature label.

  • Add OneHotLabelEncoder label based encoder.

  • Add OneHotLabelEncodingProcessor class.

  • Add DBR_Dataset class.

  • Add map_events method to MetaDataContainer to map multiple event labels into single target event label.

  • Add event_inactivity method to MetaDataContainer to get inactivity segments between events.

  • Add __version__ variable to the module.

  • Add check_installation function to check module installation.

  • Add TUTAcousticScenes_2017_FeaturesSet dataset class.

  • Add check_metadata method to dataset classes to double check meta and cross-validation setups automatically during the dataset initialization.

Updates

  • Update ProcessingChain to verify that all items in the chain are instances of Processor class.

  • Update ProbabilityItem to have index property.

  • Update ProbabilityContainer to support pickle saving and loading.

  • Update ProbabilityContainer to have as_matrix method.

  • Update majority_vote method in DecisionEncoder to be more generic (works with both labels and class IDs).

  • Move processor classes related to encoding into separate file.

  • Update load method in MetaDataContainer to translate between decimal comma and point.

  • Update data_collector function to be more generic.

  • Update formatted_value method in FancyStringifier to support fixed length strings (stf).

  • Refactor SubmissionChecker to be more flexible.

  • Update DCASEAppParameterContainer to support secondary data processing chain.

  • Update create_sequential_model function to return optionally functional API Keras model instead of default Keras sequential model.

  • Update ProgressLoggerCallback to print estimate of the remaining model learning time.

Bug fixes

  • Fix dataset class when no remote_file is set

v0.2.4

New features

  • Add TUTUrbanAcousticScenes_2018_EvaluationSet and TUTUrbanAcousticScenes_2018_Mobile_EvaluationSet dataset classes.

  • Add DCASE2018_Task5_EvaluationSet dataset class.

Updates

  • Update formatted_value method in FancyStringifier to have full coverage of float formats (float precision from 1 to 4).

Bug fixes

  • Fix TUTRareSoundEvents_2017_EvaluationSet dataset class to have correct audio path.

v0.2.3

New features

  • Add AudioWritingProcessor and MonoAudioWritingProcessor processor classes.

  • Add FeatureWritingProcessor and RepositoryFeatureWritingProcessor processor classes.

Bug fixes

  • Fix DataRepository not to have internal variables in the __dict__ after loading container from disk.

v0.2.2

In this version external dependencies of this module are minimized. External modules required for non-core functionality is not anymore included in the setup.py, and not automatically installed. Once user uses functionality requiring these rarely used external modules and module is not found, ImportError is raised with instructions to install correct module through pip. All module requirements are still available in requirements.txt.

New features

  • Add unique_source_labels property to MetaDataContainer.

  • Add file_format parameter to load and save method for ListContainer and DictContainer to force specific file format.

  • Add label_list parameter to ManyHotEncodingProcessor.

  • Add DatasetPacker class to make DCASE styled dataset packages.

  • Add dataset_exists helper function to check Dataset classes.

  • Add multi-channel audio example audio_container_ch4.

  • Add TUTUrbanAcousticScenes_2018_LeaderboardSet and TUTUrbanAcousticScenes_2018_Mobile_LeaderboardSet dataset classes.

Updates

  • Update Dataset class handle also non-text file meta files by introducing parameter evaluation_setup_file_extension.

  • Update package list handling in Dataset to support custom package extraction parameters by extra parameter package_extract_parameters.

  • Update pad method in AudioContainer to work with multi-channel audio.

  • Update compress method to produce split packages only if size limit is met.

  • Update compress method to return package filenames.

  • Update DCASE2018_Task5_DevelopmentSet dataset.

v0.2.1

New features

  • Add md5 and bytes properties to FileMixin.

  • Add two level hierarchical balancing to validation_files_balanced method in AcousticSceneDataset.

  • Add TUTUrbanAcousticScenes_2018_DevelopmentSet and TUTUrbanAcousticScenes_2018_Mobile_DevelopmentSet datasets.

  • Add float1_ci, float2_ci, float1_ci_bracket, float2_ci_bracket, float1_percentage+ci and float2_percentage+ci value types to formatted_value method in FancyStringifier.

  • Add get_set method to AppParameterContainer.

  • Add data_collector function to collect data and meta.

Updates

  • Update debug_packages method in Dataset to provide more information.

  • Update validation subset generation methods (validation_split, validation_files_dataset, validation_files_random, and validation_files_balanced) method in Dataset, AcousticSceneDataset, SoundEventDataset, and AudioTaggingDataset to allow external processing of meta data before processing through training_meta parameter.

  • Update filter method in ListDictContainer to allow filtering based on list of values.

  • Update set_label property to MetaDataItem.

  • Update filter method in MetaDataContainer to use filter method from parent class.

  • Update example applications to use current API.

  • Update random seed setting for TensorFlow in setup_keras function.

  • Update dataset_factory to handle dataset classes defined outside dcase_util.

Bug fixes

  • Fix load_from_youtube method in AudioContainer.

  • Fix example applications to work on Windows (Python 3.6).

v0.2.0

New features

  • Add row_reset and row_sep helper methods to FancyStringifier, FancyLogger, and FancyPrinter classes.

Updates

  • Update download method in RemoteFile to be more robust when encounter SSL problems.

  • Update AppParameterContainer to handle FEATURE_PROCESSING_CHAIN, DATA_PROCESSING_CHAIN.

  • Update filter method in MetaDataContainer to accept source_label and source_label_list parameters.

  • Update DCASE2018_Task5_DevelopmentSet.

Bug fixes

  • Fix construct_path method in ApplicationPaths to work in Windows as well.

  • Fix path creation in AppParameterContainer.

v0.1.9

New features

  • Add new processors FeatureReadingProcessor, DataShapingProcessor, RepositoryAggregationProcessor, RepositorySequencingProcessor, and RepositoryToMatrixProcessor.

  • Add extract method to SpectralFeatureExtractor.

  • Add automatic conversion of numeric fields when loading CSV data to ListDictContainer.

  • Add filter and get_field_unique methods to ListDictContainer.

  • Add MP4 to valid audio formats for AudioContainer.

  • Add general path modification method (Path.modify).

  • Add Keras profile cuda0_fast.

  • Add Keras utility to create optimizer instance (create_optimizer).

  • Add DCASE2018_Task5_DevelopmentSet and DCASE2013_Scenes_EvaluationSet datasets.

  • Add DataMatrix4DContainer.

  • Add plot` method to ``DataMatrix3DContainer.

  • Add support for a new annotation format for tags [filename][tab][tags] in MetaDataContainer.

  • Add zero padding to Sequencer.

  • Add header field override in load method of MetaDataContainer.

  • Add support for new compressed audio formats (OGG, MP3) in AudioContainer.

  • Add segments method in AudioContainer to split signal into non-overlapping segments with optionally skipped regions.

  • Add pad method in AudioContainer to pad signal into given length.

  • Add compress method in PackageMixin.

  • Add Package class to handle local compressed file packages.

  • Add change_axis method to DataMatrix2DContainer, DataMatrix3DContainer, and DataMatrix4DContainer.

  • Add KerasDataSequence class for data generation through processing chain.

  • Add support for data and meta processing chains to DCASEAppParameterContainer.

  • Add many_hot method in DecisionEncoder.

Updates

  • Update TUTRareSoundEvents_2017_DevelopmentSet and TUTRareSoundEvents_2017_EvaluationSet datasets.

  • Update Keras utility model_summary_string to use by default standard method from Keras.

  • Update FeatureRepository API to be aligned with Container classes.

  • Update Sequencer, SequencingProcessor, and RepositorySequencingProcessor API.

  • Update AppParameterContainer to allow change of active set even after process method has been called.

  • Update mechanism to store meta information about chain item when data is processed using processing chain.

Bug fixes

  • Fix save method in MetaDataContainer when saving with tags in CSV format.

  • Fix many methods to give more appropriate error messages.

API changes and compatibility

  • Sequencer, SequencingProcessor, and RepositorySequencingProcessor API changes:

    • frames changed to sequence_length

    • hop_length_frames to hop_length

    • padding parameter accepts now strings (zero and repeat)

v0.1.8

New features

  • Add new formats for MetaDataContainer (cpickle, CSV).

  • Add forced file formats when reading and saving containers.

  • Add Keras setup function.

  • Add frame splitting method into AudioContainer.

Bug fixes

  • Fix unicode string support when printing container information.

  • Fix data contamination through data references while manipulating data.

  • Some minor bug fixes.

v0.1.7

New features

  • Add intersection method for MetaDataContainer.

Updates

  • Update dataset class API (e.g. copy returned metadata prevent accidental manipulation, uniform method names).

Bug fixes

  • Fix data sequencing when overlapping sequencing is used.

  • Fix datasets CHiMEHome_DomesticAudioTag_DevelopmentSet, TUTAcousticScenes_2017_EvaluationSet, and TUTSoundEvents_2017_EvaluationSet.

v0.1.6

New features

  • Add CHiMEHome_DomesticAudioTag_EvaluationSet dataset.

Updates

  • Update example audio to be 16-bit audio file in wav-format instead of FLAC used earlier.

  • Update ProbabilityContainer API to be more compatible with MetaDataContainer.

  • Update MetaDataItem to be compatible with field naming used previously in DCASE baseline systems.

  • Update ui utilities.

Bug fixes

  • Fix audio reading when target sampling rate is not set.

  • Some minor bug fixes.

v0.1.5

  • Fixing PYPI package.

v0.1.4

  • Release first PYPI package.

v0.1.0

  • Initial public release.